Looks like 'Fruits' and 'Sour' has a strong correlation I didn't notice before.. I will try to create a regression model to predict sourness:

So the r-squared to beat is.. actually pretty bad. Should be easy to improve on this model.

The r-squared is positive! Okay, that is okay for only using one variable, and it is an improvement from the baseline.

I can improve on this model with more variables, but first let's see that trendline:

Let's see if adding 'Astringency' can help predict level of 'Sour' (sourness):

Hey, 60-65%, that's an improvement from a negative R-squared.

Let's see if any of the other variables with less hints of correlation can improve the model:

Not much of an imrovement there. Let's try adding 'Malty', Astringency', and 'review_appearance':

This seems about the most accurate prediction model I will get ( ~ 65-69% correlation) for 'Sour' with the LinearRegression without changing variables, or by using other tuning.